On Oct 30, 2026, at 13:21, Robert Edmonds <[email protected]> wrote:
> 
> Paul Hoffman wrote:
>> On Oct 28, 2026, at 09:28, Ray Bellis <[email protected]> wrote:
>>> 
>>> On 28/10/2026 15:57, Ben Schwartz wrote:
>>> 
>>>> A BINARY type would lose this ability, and encourage arbitrary opaque
>>>> data in DNS.  Perhaps this is what we want, but it is a big change
>>>> with some obvious downsides related to debugging and inspection.
>>>> Personally, I would be more comfortable with a "CBOR" RR type or
>>>> similar, as this at least offers the possibility of human-readable
>>>> visualization and editing.
>>> 
>>> I can see no rationale for constructing an RRtype that can *only* be used 
>>> for CBOR.  Would the DNS layer be expected to perform checks that the 
>>> content is valid CBOR?
>> 
>> This argument seems reductive. The "DNS layer" is not responsible for 
>> checking that the fields in an SOA record are valid, and yet SOA (and MX and 
>> ...) were all defined in the original DNS specification.
>> 
>> A rationale for requiring the BLOB-esque type to be CBOR is you don't need 
>> to have any other subtyping. CBOR tags are first-class data items, and a tag 
>> inside the CBOR can define what the data object is.
> 
> I'm not sure what is reductive about Ray's argument. Actually, I have
> the same question. Would the "DNS layer" be expected to validate the
> CBOR content of the record data in a hypothetical CBOR RR?

No.

> Specifically,
> would a primary nameserver need to verify that the CBOR RR's data is
> valid, well-formed CBOR data when loading RR's from a data source, such
> as a zone file? Would secondary nameservers need to verify that CBOR RR
> data in an incoming zone transfer consists of valid, well-formed CBOR
> data (and what would they do with the zone transfer if it fails such a
> check)? Would recursive resolvers need to parse CBOR RR data that they
> receive from an authoritative nameserver to ensure that it is valid,
> well-formed CBOR data before passing it on to a client?

See above.

> I'm also not sure what you mean by saying that the "DNS layer" is not
> responsible for checking that the fields in an SOA record are valid.
> It is; SOA is a well-known record type that contains two compressible
> domain names (the MNAME and RNAME fields). Receiving servers must
> decompress those fields to ensure that they do not inadvertently
> generate an invalidly formatted response when passing the SOA record
> along.

I see nothing in RFC 1035 that says that. But closer to your questions above: 
would an authoritative server reading a zone from file be expected to ensure 
that the fields in a DNSKEY record make sense cryptographically? I believe the 
answer is "no": only the validator does that.

> 
>> The DNS community is allowed to ignore IETF data format standards (in this 
>> case, a full standard), but in this case I propose that it does not.
> 
> I notice that the IETF has standardized other data formats as well, for
> instance: XML, JSON. Do we need XML and JSON RRTYPEs as well?

I didn't say that we "need" a CBOR RRtype, just that a BLOB type that has an 
identifier field and a contents field would be better off as CBOR because 
managing the identifier field will be a mess, and that mess has already been 
solved in CBOR. 

--Paul Hoffman
_______________________________________________
DNSOP mailing list -- [email protected]
To unsubscribe send an email to [email protected]

Reply via email to